The Agent Skills Directory

[PROMPT_INJECTION]: The skill describes a workflow that ingests data from external project files, creating a potential surface for indirect prompt injection attacks.
Ingestion points: Instructions in SKILL.md direct the agent to read .codestable/attention.md, .codestable/lessons/, and various legacy knowledge directories.
Boundary markers: The instructions do not define boundary markers or explicit guidelines to ignore instructions that may be embedded within the ingested files.
Capability inventory: The skill possesses the capability to modify repository code, write session state files, and spawn new agent instances for code review as described in SKILL.md.
Sanitization: No sanitization or validation logic is defined for the content retrieved from the project metadata files.
[COMMAND_EXECUTION]: The skill involves the creation and management of sub-agents and the invocation of local CLI tools for auditing refactored code. These operations are presented as part of a controlled code-review process and include instructions to avoid insecure practices like relying solely on the system PATH.

cs-refactor